Most people who bet on the NBA are picking sides. They watch a game, form an opinion, and put money on the team or the player they think will win or hit the over. The line is just the price of admission.
Value bettors do something different. They don't start with an opinion about who will win. They start with a probability — a number between zero and one that says how likely something is to happen — and then they look at the sportsbook line to see whether the price is wrong. If the price is wrong in their favour, they bet. If it isn't, they pass. Most nights, most markets, they pass.
This guide explains NBA value betting from first principles: what it is, why it works, how to do it on player props specifically, and how the Statz model surfaces value bets every day. By the end you will understand the full pipeline from raw player data to a +EV bet on tonight's slate, and you will know exactly what to do with that knowledge.
If you want to skip the theory and see today's +EV NBA props, head straight to our daily NBA value bets page. The rest of this guide is for understanding why those numbers are there.
What is value betting?
A bet has positive expected value — "+EV" — when the true probability of it winning is higher than the probability the sportsbook's price implies. That sentence is doing a lot of work, so let's break it apart.
Implied probability
Every sportsbook line is a probability in disguise. When Bet365 offers Jayson Tatum over 26.5 points at –10 (decimal 1.91), they are saying, in effect, that they price the over at roughly 52.4% likely. The number 52.4% comes from the price itself: 1 / 1.91 = 0.524.
That implied probability includes the sportsbook's vig — their margin. The over and under at –10 each imply 52.4%, which sums to 104.8% rather than 100%. The extra 4.8% is the book's edge, baked into the price.
True probability
True probability is the actual likelihood the bet will win, independent of the sportsbook. Nobody knows the true probability with perfect accuracy — not us, not the books — but with enough data and the right model you can estimate it well enough that your estimate beats the sportsbook's estimate often enough to be profitable.
If your model says Tatum has a 60% chance of going over 26.5 points and the line is implying 52.4%, you have an edge. The bet is +EV. Over a large number of similar bets at this kind of edge, you will make money. Not on every individual bet — variance is brutal in the short term — but on the long-run average.
The bet vs the outcome
This is where most casual bettors get value betting wrong. A +EV bet that loses is still a +EV bet. A −EV bet that wins is still a −EV bet. Whether you should have made the bet has nothing to do with whether it won. It has everything to do with the price you got and the true probability at the time you bet.
The mental discipline to evaluate bets by their expected value rather than their result is the single hardest part of value betting. We will come back to this when we talk about bankroll management.
Why does value betting work in the NBA?
Sportsbooks are not stupid. They employ quantitative analysts and use sophisticated models. So why is there any edge to be found at all?
Three structural reasons — specific to NBA player props — explain why a careful model can consistently beat the closing line on a meaningful subset of bets:
Player props are a high-volume, low-margin market for sportsbooks. There are roughly 200–400 player prop lines posted per night during the NBA regular season. Books cannot afford to model each one with the same care they apply to a Super Bowl moneyline. Many lines are set algorithmically off public projections, then nudged based on betting action. This leaves systematic gaps.
Public bias is asymmetric. Casual bettors love overs on stars. They love anytime touchdown scorers in the NFL and they love over-points on LeBron, Curry, and Tatum in the NBA. This bias gets priced into lines. A model that treats every player neutrally finds value on the under side of high-profile stars and on the over side of unloved role players — exactly the kinds of bets the public ignores.
Information moves slowly into player prop markets. Injury news, rotation changes, blowout potential, back-to-back fatigue, defensive matchup data — some of this gets priced in instantly, but a lot of it takes hours, and some of it is never priced in properly because the books rely on stale priors. A model that ingests fresh data faster than the books update their lines has a structural advantage.
None of this means value bets are easy to find. Most lines are priced correctly. The job is to find the small subset that aren't — and to do it consistently, without fooling yourself.
How to calculate the edge on an NBA prop bet
The full mechanics of expected value calculation deserve their own treatment, and we wrote one: see our guide on how to calculate EV on NBA prop bets for the formulas and worked examples. The short version follows.
The four numbers you need
Your projection: a point estimate of the player's stat (e.g. 28.4 points)
The line: the sportsbook number (e.g. over 26.5 points)
The price: the decimal odds offered (e.g. 1.91)
Your distribution: how confident is your projection? A projection of 28.4 points with a standard deviation of 6 points implies a different over-probability than the same projection with a standard deviation of 3 points.
The calculation
Expressed simply: convert your projection plus your distribution into a probability that the over hits, then convert the sportsbook price into its implied probability. The difference is your edge. Edge expressed as a percentage of the sportsbook's implied probability is the metric we use on the value bets page.
Example. Tatum projection: 28.4 points, distribution implying a 60% chance of clearing 26.5. Bet365 line: over 26.5 at 1.91, implying 52.4%. Edge: (60% − 52.4%) / 52.4% = +14.5%. That's the number you'd see in the right-hand column of our value bets table.
A +14.5% edge does not mean you will win 14.5% more often than the line suggests. It means your model thinks you will. Your model can be wrong — about Tatum's projection, about the distribution, about the matchup. The edge is a hypothesis, not a fact.
Where do projections come from?
A projection is the model's best estimate of how a player will perform tonight. Good projections combine three layers.
Baseline form
How has the player performed this season, weighted toward recent games and adjusted for pace? A player on a fast-paced team running 105 possessions per game produces more counting stats per minute than the same player on a slow team. Pace adjustment is non-negotiable.
Matchup
Who is the opponent and how do they defend the player's position? A point guard facing a team that allows the most points to point guards in the league should be projected higher than the same point guard facing the best defensive team. This is what "defense vs position" data captures.
Context
Is the game expected to be a blowout (rest implications)? Is the player on a back-to-back? Are key teammates out, changing the usage distribution? Is the line for tonight set at home or on the road? Each of these moves the projection by a meaningful amount.
The Statz model combines these layers using a Ridge Regression baseline (which captures form and pace) and a Gradient Boosting layer (which captures the non-linear matchup and context interactions). The two are stacked into a single ensemble projection, then converted into a probability distribution over outcomes. For a deeper dive into the modelling approach, see our companion piece on NBA player prop projections vs sportsbook lines.
How to use a daily value bets list
Looking at a list of +EV bets and just betting all of them is one strategy. It's not a bad strategy, but it's not the best one either. Here is how serious value bettors actually use a list like the one on statz.ai.
Filter by edge threshold
Lower-edge bets are more likely to be model noise. Most quantitative bettors apply a filter — 5%, 10%, sometimes 15% — below which they don't bet. The right threshold depends on how much you trust the model and how much variance you can stomach.
Filter by stat type
Models are not equally accurate across stat types. Points and rebounds are easier to project than assists, which depend heavily on team context, or threes, which are high-variance. Track your hit rate by stat type and adjust your filter accordingly.
Cross-check the news
Models lag breaking news by minutes to hours. If a value bet is showing on a player whose status is in doubt, check the latest beat reporter on Twitter before you click. The line might already be moving.
Stake by Kelly
Bet sizing matters more than bet selection. The Kelly Criterion gives you the mathematically optimal stake for a given edge and bankroll. Most bettors should use fractional Kelly — typically 25–50% of full Kelly — to soften the variance. We cover this in detail in our bankroll management guide for +EV NBA bettors.
Common mistakes
Chasing the highest edge
The +50% edge bet at the top of the list is not always the best bet on the list. A massive edge often means the model is missing something the market knows — an injury, a rotation change, a recent trade. Investigate before betting; don't just stake the top of the table blindly.
Ignoring closing line value
The most rigorous test of whether you're truly +EV is closing line value (CLV). Did you bet at a better price than the line closed at? Bettors who consistently beat the closing line are profitable in the long run, almost without exception. If your bets are consistently behind the closing line, the model is the problem, not your luck.
Confusing variance with skill
Hitting 60% over a weekend doesn't mean your model is good. Hitting 53% over 2,000 bets at average +5% edge means your model is good. Sample size matters more than streaks.
Betting on books that limit you
Most major sportsbooks limit or close accounts that show consistent +EV behaviour. Plan for this. Spread action across multiple books, vary stake sizing, and don't bet exclusively on the highest-edge plays — mix in some lower-edge bets and the occasional recreational bet to fly under the radar. This is the unglamorous reality of value betting at scale.
How Statz helps
Statz produces an independent projection for every NBA player prop, every day, and compares it against the live Bet365 line. The output is a daily list of +EV NBA props sorted by edge percentage. Every projection is the output of the same Ridge + Gradient Boosting ensemble — transparent in methodology, consistent in approach, and tracked against actual results so you can see how the model performs over time.
If you want to start finding NBA value bets without building your own model from scratch, the daily value bets page is the simplest place to begin. From there, the four pieces below will deepen your understanding of each part of the pipeline:
How to Calculate Expected Value (EV) on NBA Prop Bets — the formulas and worked examples behind the edge column
Bet365 NBA Props: How to Find Edges in the Lines — why we benchmark against Bet365 specifically
NBA Player Prop Projections vs Sportsbook Lines: A Quant Approach — how the model is built
Bankroll Management for +EV NBA Bettors: Kelly Criterion & Beyond — how to size your bets so you don't go broke even when you're right
The bottom line
Value betting works in the NBA because player prop markets are vast, public bias is consistent, and information moves unevenly. It does not work because the bettor is smarter than the market on every bet — it works because they only bet when they believe they are. Most of the time, that means not betting.
If you take one thing from this guide, take this: the bet that loses tonight does not invalidate the process that produced it. Track your edges, track your closing line value, size your stakes properly, and let the long run do its work. That's the discipline. The rest is data.